Unsupervised experiments

`EmbeddingUnsupervisedExperiment`

This class is used to run experiments on synthetic toy functions.

EmbeddingUnsupervisedExperiment(
    task_type,
    **kwargs,
)

Parameters

Parameter	Type	Default	Description
`task_type`	—	Required	—
`**kwargs`	—	—	—

`EmbeddingUnsupervisedExperiment.plot`

View source

`EmbeddingUnsupervisedExperiment.run`

View source

`GenerateSyntheticDataExperiment`

View source

This class is used to run experiments on generating synthetic data.

GenerateSyntheticDataExperiment(
    task_type,
    **kwargs,
)

Parameters

Parameter	Type	Default	Description
`task_type`	—	Required	—
`**kwargs`	—	—	—

`GenerateSyntheticDataExperiment.plot`

View source

`GenerateSyntheticDataExperiment.run`

View source

Generate synthetic data and store it on the experiment instance. The following properties are also set

data_real: input X data as a DataFrame, potentially resampled
data_synthetic: synthetic_X as a DataFrame, potentially resampled
data: data_real and data_synthetic concatenated If one of data_real or data_synthetic has fewer rows, it is resampled with replacement so both have max(n_input_samples, n_samples) rows. data_real, data_synthetic, and data have an additional real_or_synthetic column that indicates if the data is real or synthetic.

GenerateSyntheticDataExperiment.run(
    tabpfn,
    *,
    categorical_features = None,
    should_plot = True,
    **kwargs,
)

Parameters

Parameter	Type	Default	Description
`tabpfn`	—	Required	A `TabPFNUnsupervisedModel` used to learn the joint distribution of the selected features and sample synthetic rows.
`categorical_features`	—	`None`	Column indices of `X` (same index space as `indices`) to treat as categorical. Indices not present in `indices` are ignored. Defaults to `None`, in which case the model auto-detects categorical columns at `fit` time.
`should_plot`	—	`True`	Whether to render the pairwise plot. Defaults to `True`.
`**kwargs`	—	—	Keyword arguments controlling the run: X: Input data array of shape `(n_input_samples, n_features)`. y: Targets (unused for unsupervised generation; may be empty). attribute_names: Column names for every column in `X`. indices: Column indices of `X` to model. Defaults to all columns. temp: Sampling temperature. Defaults to `1.0`. n_samples: Number of synthetic rows to generate. Defaults to `X.shape[0]`. n_permutations: Number of feature-order permutations to average. Defaults to `3`. dag: Optional causal DAG passed to the generator.

The synthetic data is stored on the following instance property

synthetic_X: array with shape (n_samples, n selected columns)

`OutlierDetectionUnsupervisedExperiment`

View source

This class is used to run experiments for outlier detection.

OutlierDetectionUnsupervisedExperiment(
    task_type,
    **kwargs,
)

Parameters

Parameter	Type	Default	Description
`task_type`	—	Required	—
`**kwargs`	—	—	—

`OutlierDetectionUnsupervisedExperiment.plot`

View source

`OutlierDetectionUnsupervisedExperiment.plot_two`

View source

`OutlierDetectionUnsupervisedExperiment.run`

View source

Estimate per-sample outlier scores for the selected features.

OutlierDetectionUnsupervisedExperiment.run(
    tabpfn,
    overwrite_baseline_cache = False,
    overwrite_tabpfn_cache = True,
    *,
    categorical_features = None,
    should_plot = True,
    **kwargs,
)

Parameters

Parameter	Type	Default	Description
`tabpfn`	—	Required	A `TabPFNUnsupervisedModel` used to estimate sample density.
`overwrite_baseline_cache`	—	`False`	Unused placeholder kept for API symmetry.
`overwrite_tabpfn_cache`	—	`True`	Unused placeholder kept for API symmetry.
`categorical_features`	—	`None`	Column indices of `X` (same index space as `indices`) to treat as categorical. Indices not present in `indices` are ignored. Defaults to `None`, in which case the model auto-detects categorical columns at `fit` time.
`should_plot`	—	`True`	Whether to render the density plot. Defaults to `True`.
`**kwargs`	—	—	Keyword arguments controlling the run: X: Input data array of shape `(n_samples, n_features)`. y: Targets (unused; may be empty). attribute_names: Column names for every column in `X`. indices: Column indices of `X` to model. Defaults to all columns. n_permutations: Number of feature-order permutations to average. Defaults to `3`.

Returns

Type	Description
—	`dict`: Mapping with key `"log_p"` holding the per-sample log-density (lower values indicate more likely outliers).

Reference

tabpfn

tabpfn-client

tabpfn-extensions

REST API

Unsupervised experiments

`EmbeddingUnsupervisedExperiment`

`EmbeddingUnsupervisedExperiment.plot`

`EmbeddingUnsupervisedExperiment.run`

`GenerateSyntheticDataExperiment`

`GenerateSyntheticDataExperiment.plot`

`GenerateSyntheticDataExperiment.run`

`OutlierDetectionUnsupervisedExperiment`

`OutlierDetectionUnsupervisedExperiment.plot`

`OutlierDetectionUnsupervisedExperiment.plot_two`

`OutlierDetectionUnsupervisedExperiment.run`

​EmbeddingUnsupervisedExperiment

​EmbeddingUnsupervisedExperiment.plot

​EmbeddingUnsupervisedExperiment.run

​GenerateSyntheticDataExperiment

​GenerateSyntheticDataExperiment.plot

​GenerateSyntheticDataExperiment.run

​OutlierDetectionUnsupervisedExperiment

​OutlierDetectionUnsupervisedExperiment.plot

​OutlierDetectionUnsupervisedExperiment.plot_two

​OutlierDetectionUnsupervisedExperiment.run

`EmbeddingUnsupervisedExperiment`

`EmbeddingUnsupervisedExperiment.plot`

`EmbeddingUnsupervisedExperiment.run`

`GenerateSyntheticDataExperiment`

`GenerateSyntheticDataExperiment.plot`

`GenerateSyntheticDataExperiment.run`

`OutlierDetectionUnsupervisedExperiment`

`OutlierDetectionUnsupervisedExperiment.plot`

`OutlierDetectionUnsupervisedExperiment.plot_two`

`OutlierDetectionUnsupervisedExperiment.run`